欢迎访问《麦类作物学报》编辑部网站

作者	单位
李岚涛，郭宇龙，韩鹏，王丹丹，王宜伦	（1.河南农业大学资源与环境学院，河南郑州 450002；2.博爱县农业农村局，河南博爱 454450）

摘要点击次数: 648

全文下载次数: 663

中文摘要:

为实现冬小麦不同生育时期地上部生物量的高光谱监测，2017-2019年分别在河南省鹤壁市、原阳县和温县布置冬小麦氮肥梯度田间试验，分别于分蘖期、拔节期、抽穗期和灌浆期测定冬小麦地上部生物量及其冠层原位高光谱反射率（400～950 nm），并采用Pearson相关分析明确两者间定量回归关系，再分别利用支持向量机（support vector machine, SVM）和偏最小二乘回归（partial least square, PLS）建立预测模型并进行精度验证，以确定最优光谱监测时期和有效波段。结果表明，冬小麦地上部生物量与冠层高光谱反射率在可见光区（400～715 nm）呈负相关，在近红外区（715～950 nm）呈正相关，且相关性表现为分蘖期<拔节期<灌浆期≤抽穗期。生育时期间模型精度差异较大，抽穗期效果最优，SVM和PLS模型的验证决定系数分别为0.877和0.859，相对分析误差分别为2.429和2.340；灌浆期次之，决定系数分别为0.835和0.830，相对分析误差分别为2.416和1.814；分蘖期最低，决定系数分别为0.693和0.750，相对分析误差分别为1.063和0.894。同时，冬小麦地上部生物量有效波段在生育时期间具有明显的异同性，分蘖期时有效波段在可见光-近红外区均有明显的均衡分布，至拔节期时产生明显的短波“蓝移”现象，抽穗期“蓝移”现象更显著，而至灌浆期则表现出明显的长波“红移”特征。此后，再次构建基于有效波段的冬小麦不同生育时期地上部生物量SVM和PLS监测模型，决定系数和相对分析误差分别高于0.72和1.40，预测精度较理想，能够满足无损和精准监测需求。

英文摘要:

In order to realize hyperspectral monitoring of shoot biomass of winter wheat at different growth stages, field experiments were conducted in three locations(Hebi City, Yuanyang County and Wenxian County of Henan Province) from 2017 to 2019. The shoot biomass and canopy hyperspectral reflectance(400-950 nm) of winter wheat were measured at tillering stage, jointing stage, heading stage and filling stage.Support vector machine(SVM) and partial least square(PLS) model were used to establish the prediction model and verify the accuracy to determine the optimal spectral monitoring period and effective band.The results showed that there was a negative correlation between shoot biomass and canopy hyperspectral reflectance in visible light region (400-715 nm)and positive correlation in near-infrared region(715-950 nm), and the correlation ranked as: tillering stage < jointing stage < filling stage ≤ heading stage. The results showed that the accuracy of the models varied greatly during the growth period, and the effect of heading stage was the best. The validation determination coefficients(r) of SVM and PLS models were 0.877 and 0.859,and relative percent deviations(RPD_val) were 2.429 and 2.340, respectively.The r of filling stage were 0.835and 0.830, and the RPD_val were 2.416 and 1.814, respectively.The poorest performance was acquired using the SVM and PLS model at tillering stage, with the r of 0.693and 0.750, and RPD_val of 1.063, 0.894, respectively. The effective wavelengths for above-ground biomass estimation were also had significant differences among the four growing stages. At tillering stage, the effective bands have an obvious and balanced distribution in the visible and near-infrared region. However, a strong “blue shift” phenomenon of the effective wavelengths compared with that of the tilling stage was observed at jointing and heading stages, and “red shift” phenomenon towards the longer wavelengths was shown at filling stage. Additionally, the newly developed SVM and PLS model using the effective wavelengths at different growth stages for shoot biomass of winter wheat prediction also performed well(r >0.72; RPD>1.40). The overall results indicate the winter wheat shoot biomass can be reliably estimated with the canopy hyperspectral methods established in this study.

查看全文查看/发表评论下载PDF阅读器

关闭